2015 IEEE Signal Processing and Signal Processing Education Workshop (SP/SPE)
SPECTRAL ANALYSIS OF STOCK-RETURN VOLATILITY, CORRELATION, AND BETA
A. Shomesh E. Chaudhuri
Massachusetts Institute of Technology
Electrical Engineering and Computer Science
Cambridge, MA
ABSTRACT
We apply spectral techniques to analyze the volatility and cor-
relation of U.S. common-stock returns across multiple time
horizons at the aggregate-market and individual-firm level.
Using the cross-periodogram to construct frequency band-
limited measures of variance, correlation and beta, we find
that volatilities and correlations change not only in magnitude
over time, but also in frequency. Factors that may be respon-
sible for these trends are proposed and their implications for
portfolio construction are explored.
Index Terms— spectral analysis, volatility, correlation,
beta, portfolio theory, financial engineering
1. INTRODUCTION
It has been observed that the volatility of securities and their
correlation with the stock market are not constant, but change
over time. This variation in volatility and correlation has
important implications for any theory of risk, return, and
portfolio construction. However, standard measures do not
distinguish between the short- and long-term components
of risk and co-movement. The fact that economic shocks
produce distinct effects on stock return dynamics at different
time horizons suggests that frequency-specific measures of
volatility and correlation may yield several new insights.
First, studying the frequency components of stock return
processes can reveal new features of the underlying economic
structure driving these processes. Second, since the time hori-
zon for investments can range from microseconds to decades,
the risks specific to these time horizons can be gauged and
taken into consideration during portfolio construction. Third,
as trading strategies become ever faster, the frequency domain
provides a convenient framework for comparing investment
processes that operate on different timescales and, more im-
portantly, for diversifying across these timescales. Finally,
∗ The views and opinions expressed in this article are those of the authors
only and do not necessarily represent the views and opinions of any other
organizations, any of their affiliates or employees, or any of the individuals
acknowledged above. Research support from the MIT Laboratory for Finan-
cial Engineering is gratefully acknowledged.
978-1-4673-9169-6/15/$31.00 ©2015 IEEE
B. Andrew W. Lo∗
Massachusetts Institute of Technology
Sloan School of Management, CSAIL, EECS
Cambridge, MA
frequency-domain measures offer surprisingly simple repre-
sentations of complex dynamics that are cyclical.
In this article, we propose several new frequency-domain
measures of stock-market risk and expected return and show
how they can complement traditional approaches to portfolio
management. We begin in Section 2 by reviewing the liter-
ature on volatility, portfolio theory, and spectral analysis as
applied to finance. In Section 3 we present the basic decom-
position of variance, correlation, and beta into their frequency
components, and in Section 4 we apply this framework to the
historical returns of U.S. stocks. We explore the applications
of these measures to portfolio theory in Section 5, and con-
clude in Section 6.
2. LITERATURE REVIEW
There is an enormous literature on statistical models for cap-
turing the time variation in volatility. Simple models such
as the rolling standard deviation used by Officer [1] have
given way to sophisticated econometric techniques such as
ARCH and GARCH models introduced by Engle [2] and
other stochastic volatility models. Partial surveys of these
methods are given by Campbell, Lo, and MacKinlay [3],
Broto and Ruiz [4], and Bauwens, Laurent and Rombouts [5].
In this article, we focus on characterizing the time-horizon-
dependent components of volatility and correlation; hence,
our use of spectral analysis.
Spectral analysis has a long history in econometrics [6],
[7], with applications ranging from business cycle analy-
sis [8] to option valuation [9]. Often the analysis relies on
the Fourier transform (e.g., [10], [11]), however recently
wavelets (e.g., [12], [13], [14]) and the Hilbert-Huang trans-
form (e.g., [15]) have also been used to study financial data
in the time-frequency domain. In this article, the cross-
periodogram—calculated using the short-time Fourier trans-
form [16]—forms the basis of our spectral analysis. The
cross-periodogram, which is the decomposition of the inner
product of two time series into their frequency components,
was introduced into the economic literature by Engle [17]
as a component of band-spectrum regression. Using a sim-
ilar insight, we use the frequency-band-limited adaptations
232
 2015 IEEE Signal Processing and Signal Processing Education Workshop (SP/SPE)

of variance, correlation, and beta, to study the statistical More generally, averaging over the cross-periodogram
properties of asset returns at multiple time horizons. This can be used decompose the sample covariance into its fre-
framework connects spectral analysis to the standard tools quency components:
of modern portfolio theory developed by Markowitz [18], al-
T −1
lowing us to analyze the time-horizon properties of portfolio 1 X ∗
cov(ri , rj ) = Ri,k Rj,k . (5)
construction. T2
k=1

This technique uses the DFT to express the returns in the fre-
3. SPECTRAL ANALYSIS quency domain and then analyzes their phase. When the re-
turns are in phase at a given frequency, the contribution that
We use the discrete Fourier transform (DFT) to decompose frequency makes to the overall covariance is positive. When
the covariance of returns into its frequency components, from they are out of phase, then that particular frequency’s contri-
which we calculate time-horizon-specific correlations and be- bution will be negative. Values of k that are symmetric about
tas. T /2 (e.g., k = 1 and k = T −1) have the same frequency and
Let ri,t be the one-period return of stock i between dates their contributions to the sample variance are complex conju-
t−1 and t. The sample variance of returns over a interval from gates. Therefore, the contribution from a single frequency or
t = 0, . . . , T −1 can then be calculated as: subset of frequencies will always be real.
We can calculate the contribution of a subset of frequen-
T −1
1 X cies, K ⊆ {0, . . . , T−1}, to the sample covariance by limiting
var(ri ) = (ri,t − ri )2 , (1)
T t=0 the summation in (5) to the frequencies of interest:2 ,
1 X ∗
where ri is the sample mean of returns, covK (ri , rj ) = Ri,k Rj,k . (6)
T2
k∈K
T −1
1 X This corresponds to calculating the sample covariance of the
ri = ri,t . (2)
T t=0 inverse DFT reconstruction of returns, restricted to the speci-
fied frequency subset.
This calculation is exactly equivalent to the one formed by A few important implementation details still remain. Win-
averaging over the periodogram: dowing procedures (e.g., multiplication by a Hamming win-
dow) can be applied to the data before taking the Fourier
T −1 transform used to calculate the modified periodogram. This
1 X ∗
var(ri ) = Ri,k Ri,k (3) procedure will generally decrease spectral leakage at the ex-
T2
k=1 pense of reducing spectral resolution. Moreover, averaging
cross-periodograms formed over overlapping time intervals
where Ri,k are the T -point DFT coefficients1 of ri,t : can reduce the variance of the spectral estimates at the ex-
pense of increased bias. This method is known as the Bartlett
T −1
X 2πkt method if no window other than a rectangular window is ap-
Ri,k = ri,t e−j T , 0 ≤ k ≤ T −1 . (4)
plied to the data sections, and as Welch’s method if a tapered
t=0
window is applied. We refer the reader to Oppenheim and
Note that k = 0, the zero frequency, is not involved in (3) Schafer [16, Chapter 10] for a more detailed discussion of
since adding or subtracting a constant to the time series ri,t these advanced implementation techniques.
does not change its sample variance. In this form, the con- This frequency-domain analysis can be extended to time-
tribution to the sample variance by the kth frequency, where horizon specific calculations of the beta and correlation co-
k ∈ {1, . . . , T − 1}, is real valued and clearly visible. The efficients. The frequency-restricted beta coefficient, βK , and
lowest non-zero frequency occurs at k = 1 and the highest correlation coefficient, ρK , can be calculated as:
frequency occurs at the value of k closest to T /2. Values of  1/2
k that are symmetric about T /2 (e.g., k = 1 and k = T − 1) varK (ri )
βK (ri , rj ) = ρK (ri , rj ) (7)
have the same frequency and their contributions to the sample varK (rj )
variance are the same. The relation h = T Ts /k, where Ts is
where
the time between samples and 0 ≤ k ≤ T /2, can be used to
convert the kth frequency to its corresponding time horizon. covK (ri , rj )
ρK (ri , rm ) = . (8)
[varK (ri )varK (rj )]1/2
1 These coefficients can easily be calculated using the algorithms pro-
vided by any standard statistical software program. For example, we used 2 Note that pairs of elements that correspond to the same frequency should

the “fft” function in MATLAB. be included together in K.

233
 2015 IEEE Signal Processing and Signal Processing Education Workshop (SP/SPE)
Since returns are real-valued, −1 ≤ ρK (ri , rj ) ≤ 1, and (8) is
computationally equivalent to calculating the sample correla-
tion of the inverse DFT reconstructions of returns, restricted
to the frequencies specified by K. Engle [17] shows how the
band spectrum regression coefficient βK can be generalized
to the case of multiple factors.
4. MEASURING TRENDS IN VOLATILITY AND
CORRELATION
Our empirical analysis focuses on all stocks in the University
of Chicago’s CRSP Database from May 10, 1963 to Decem-
ber 31, 2014. Specifically, we use only U.S. common stocks
with CRSP share code 10 and 11, which eliminates REIT’s,
ADR’s, and other non-common-stock securities. CRSP occa-
sionally reports returns and prices that are based on bid-ask
quote midpoints (for infrequently traded securities); we elim-
inate a stock from our sample if more than 5% of its prices
are based on these values in any windowed time interval.
In Fig. 1 we show the trailing 750-trading-day annualized
variance and variance spectrum of the CRSP value-weighted
market index. We compute the annualized variance based on
daily data. The figure shows large spikes in volatility across
all frequencies during the oil shocks of the 1970s, the stock
market crash of 1987, the Tech Bubble, and the Financial Cri-
sis of 2007–2008. Moreover, the variance spectra over the
past two decades show more high-frequency content than any
of the earlier periods, which is not surprising considering the
ever increasing speed and volume of trading.
Figs. 2 and 3 show the cross-sectional average of the
trailing 750-trading-day annualized variance and correlation
spectra when applied to stocks sorted into NASDAQ and
NYSE/AMEX subsets by their CRSP exchange code identi-
fier. We find that, throughout the 1990s, the average variance
of NASDAQ securities trended upwards, especially at the
shorter time horizons (between 1 day and 1 week). In con-
trast, there is no discernible trend in the average variance of
stocks trading on the NYSE/AMEX during this period. Since
this upward trend in short-run individual stock volatility did
not translate into increased market volatility, we see in Fig. 2
that, as expected, the average correlation at these high fre-
quencies decrease. We also note that the correlation spectra in
both Figs. 2 and 3 show large increases in the average corre-
lation with the market across all frequencies during the stock
market crash of 1987 and the Financial Crisis of 2007–2008.
Moreover, correlations with the market have tended to trend
upward over the past two decades.
It is natural to ask what might explain the increasing
volatility and decreasing market correlation observed in these
NASDAQ securities during the 1990s. We speculate that
this trend may be partly explained by the high-frequency dy-
namics induced by the advent and proliferation of electronic
trading networks during this period. NASDAQ embraced
electronic trading much earlier than the NYSE/AMEX and
234
0.1
Variance
0.05
0
70 80 90 00 10
2 0.01
Horizon [day]
Variance
3
6
30
0
70 80 90 00 10
Year
Fig. 1. The top panel shows the trailing 750-trading-day an-
nualized variance of the value-weighted market index pro-
vided in the CRSP data set between May 10, 1963 and De-
cember 31, 2014. The bottom panel shows the corresponding
variance spectrum partitioned into 15 frequency bands.
provided various electronic platforms for both market mak-
ers and customers. The algorithmic trading strategies em-
ployed on these electronic platforms may have increased the
amount of noise at these shorter time horizons, while re-
maining uncorrelated with general market movements. These
trends would have been particularly noticeable in stocks
with low baseline prices and small market capitalizations—
such as those found predominantly on the NASDAQ stock
exchange—since high-frequency price dynamics on the order
of the bid-ask spread would have subsequently larger effects
on returns.
An interesting area for further research is to determine the
underlying processes driving these volatility and correlation
measures, and, in particular, explaining the upward trend in
high-frequency volatility in the 1990s.
5. IMPLICATIONS FOR PORTFOLIO
CONSTRUCTION
We have shown that individual stock volatility and correla-
tions change over time and across frequencies. An implica-
tion of this finding is that a portfolio can be designed to max-
imize expected returns while minimizing its risk over a given
frequency band. Since investors have different time horizons,
it is rational for them to seek to minimize the risk specific to
a particular frequency band. For example, short-run volatil-
ity, even if correlated with their portfolio, may not be impor-
tant to their investment goals if their time horizon is much
longer. Similarly, low-frequency power in returns and corre-
lations may be insignificant to a high-frequency trader who
does not operate on the same timescale. Our frequency de-
compositions provide a systematic framework for performing
such an analysis.
In mean-variance portfolio theory, given a target value
 2015 IEEE Signal Processing and Signal Processing Education Workshop (SP/SPE)

1 0.5 0.5 0.8

Correlation
Correlation

Variance
Variance

0.5 0.25 0.25 0.4

0 0 0 0
70 80 90 00 10 70 80 90 00 10
2 0.08 2 0.04

Horizon [day]
Horizon [day]

Variance
Variance
3 3
0.04 0.02
6 6
30 30
0 0
70 80 90 00 10 70 80 90 00 10
2 0.6 2 0.8

Horizon [day]
Horizon [day]

Correlation
Correlation
3 3
0.3 0.4
6 6
30 30
0 0
70 80 90 00 10 70 80 90 00 10
Year Year

Fig. 2. The top panel shows the cross-sectional average for
NASDAQ securities of their trailing 750-trading-day return
annualized variance and correlation with the value-weighted
market index return between April 13, 1983 and December
31, 2014. The middle and bottom panels show the corre-
sponding average annualized variance and correlation spectra,
respectively, partitioned into 15 frequency bands.
Fig. 3. The top panel shows the cross-sectional average
for NYSE/AMEX securities of their trailing 750-trading-day
return annualized variance and correlation with the value-
weighted market index return between May 10, 1963 and De-
cember 31, 2014. The middle and bottom panels show the
corresponding average annualized variance and correlation
spectra, respectively, partitioned into 15 frequency bands.

(µ̃) for the expected portfolio return, the efficient portfolio

ratio of the variance of portfolio B to that of portfolio A is
weights (w̃) are those that minimize the portfolio variance for
approximately:
all portfolios with expected return µ̃. Mathematically, the op-
timization problem can be written as,
var(rB )
≈ 1 + 2δ(β(ri , rA ) − 1) (11)
w̃ = arg min wT Σw (9) var(rA )
w

subject to the constraints
wT µ = µ̃ and wT 1 = 1
where wi is the portfolio weight on the ith security, µi =
E[ri ] and Σi,j = cov(ri , rj ).
The inputs for the optimization problem are, therefore,
the expected returns and covariance matrix of these securi-
ties. As a first-order approximation, sample estimates can be
used for these values. Accordingly, a time-horizon-specific
mean-variance optimization, restricted to the frequency band
K, can be developed by simply replacing the sample standard
deviation and correlation matrix estimates with those based
on (6) and (8), respectively. This framework has the attractive
feature that the optimization techniques developed to solve
for the efficient frontier are still valid since the form of inputs
are not affected.
Now suppose that an investor currently holds portfolio A,
and wishes to shift a small fraction of his portfolio weight, δ,
to security i to create portfolio B. It can be shown that the
which implies that if a small amount of security i is added
to portfolio A, the variance of the portfolio will increase if
(10) β(ri , rA ) > 1, and will decrease if β(ri , rA ) < 1. Taking
advantage of the linear properties of the Fourier transform,
we find that:
varK (rB )
≈ 1 + 2δ(βK (ri , rA ) − 1) (12)
varK (rA )
which provides a similarly intuitive interpretation, in terms of
portfolio construction, for the restricted-frequency beta coef-
ficient, βK .
These simple examples highlight the versatility of covK ,
ρK and βK . As the frequency band-limited counterparts to
covariance, correlation, and beta, their applications are not
limited to modern portfolio theory, but can be applied to al-
most any theory of risk, reward, and portfolio construction.
An interesting area for future research is to investigate the the
practical advantages of such a framework.

235
 2015 IEEE Signal Processing and Signal Processing Education Workshop (SP/SPE)

6. CONCLUSION [10] C. Croux, M. Forni, and L. Reichlin, “A measure of

comovement for economic variables: Theory and em-
In this article, we apply spectral analysis to characterize the pirics,” Review of Economics and Statistics, vol. 83, pp.
behavior of stock-return volatility, correlation, and beta across 232–241, May 2001.
multiple time horizons at the individual-firm and aggregate-
market level. Our approach exploits a frequency-band analy- [11] J. Breitung and B. Candelon, “Testing for short-
sis of variance and correlation to analyze the daily returns of and long-run causality: A frequency-domain approach,”
securities trading on the NASDAQ and NYSE/AMEX stock Journal of Econometrics, vol. 132, pp. 363–378, June
exchanges. We identified periods of increasing variance and 2006.
decreasing correlation in NASDAQ securities at the shorter
[12] J.B. Ramsey, “Wavelets in economics and finance: Past
time horizons between 1 day and 1 week. We speculate that
and future,” Studies in Nonlinear Dynamics & Econo-
this trend may result in part from changes induced by elec-
metrics, vol. 6, pp. 1–27, 2002.
tronic trading. Finally, we suggest some potential applica-
tions for the frequency band-limited adaptations of covari- [13] A. Rua, “Measuring comovement in the time-frequency
ance, correlation, and beta to portfolio theory. These con- space,” Journal of Macroeconomics, vol. 32, pp. 685–
siderations are particularly useful when portfolio goals differ 691, June 2010.
across time horizons, and when investors wish to target spe-
cific horizons because of their preferences and life cycle. [14] A. Rua, “Wavelets in economics,” Economic Bul-
letin and Financial Stability Report Articles, pp. 71–79,
2012.
7. REFERENCES
[15] N.E. Huang, M.L. Wu, W. Qu, S.R. Long, S.S.P. Shen,
[1] R.R. Officer, “The variability of the market factor of the and J.E. Zhang, “Applications of Hilbert-Huang trans-
New York Stock Exchange,” Journal of Business, vol. form to non-stationary financial time series analysis,”
46, pp. 434–453, July 1973. Applied stochastic models in business and industry, vol.
[2] R.F. Engle, “Autoregressive conditional heteroscedas- 19, pp. 245–268, 2003.
ticity with estimates of the variance of United Kingdom
[16] A.V. Oppenheim and R.W. Schafer, Discrete-Time Sig-
inflation,” Econometrica, vol. 50, pp. 987–1007, July
nal Processing, Prentice Hall, 3rd edition, 2009.
1982.
[17] R.F. Engle, “Band spectrum regression,” International
[3] J.Y. Campbell, A.W. Lo, and A.C. MacKinlay, The
Economic Review, vol. 15, pp. 1–11, February 1974.
Econometrics of Financial Markets, Princeton Univer-
sity Press, 1997. [18] H. Markowitz, “Portfolio selection,” The Journal of
[4] C. Broto and E. Ruiz, “Estimation methods for stochas- Finance, vol. 7, pp. 77–91, March 1952.
tic volatility models: A survey,” Journal of Economic
Surveys, vol. 18, pp. 613–649, December 2004.
[5] L. Bauwens, S. Laurent, and J.V.K. Rombouts, “Multi-
variate GARCH models: a survey,” Journal of Applied
Econometrics, vol. 21, pp. 79–109, 2006.
[6] C.W.J. Granger and M. Hatanaka, Spectral Analysis
of Economic Time Series, Princeton University Press,
1964.
[7] C.W.J. Granger and R.F. Engle, “Applications of spec-
tral analysis in econometrics,” Handbook of Statistics,
vol. 3, pp. 93–109, 1983.
[8] M. Baxter and R.G. King, “Measuring business cy-
cles: Approximate band-pass filters for economic time
series,” review of economics and statistics, vol. 81, pp.
575–593, November 1999.
[9] P. Carr and D. Madan, “Option valuation using the fast
fourier transform,” Journal of computational finance,
vol. 2, pp. 61–73, 1999.

236